Regression Models for Estimating Kinematic Gait Parameters with Instrumented Footwear
Regression Models for Estimating Kinematic Gait Parameters with Instrumented Footwear
复制标题
使用仪器鞋估计运动学步态参数的回归模型
DOI：
10.1109/biorob.2018.8487972
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
D. Zanotto
中科院分区：
文献类型：
--
作者：
Huanghe Zhang;Mey Olivares Tay;Zeynep Suar;M. Kurt;D. Zanotto
Quantitative gait assessment typically involves optical motion capture systems and force plates, which result in high operating costs. Footwear-based motion tracking systems can provide a portable and affordable solution for real-time gait analysis in unconstrained environments. However, the relatively low accuracy of these systems still represents a barrier to their widespread use. In this paper, we show that linear and learning-based regression models can substantially improve the raw estimates of a set of kinematic gait parameters obtained with instrumented insoles (SportSole) from a group of $\mathbf{N}=\pmb{9}$ healthy subjects who walked at different speeds. Least Absolute Shrinkage and Selection Operator (LASSO) and Support Vector Regression (SVR) models are compared in terms of accuracy, precision, and robustness to change in gait speed, using gold-standard equipment to generate reference data. Results indicate that SVR is superior to LASSO. Indeed, the mean absolute errors (MAE) in stride length, velocity and foot-ground clearance were $\pmb{1.28}\pm\pmb{0.19\%}. \pmb{1.62}\pm\pmb{0.42\%}$ and $\pmb{3.72}\pm \pmb{0.87\%}$ for LASSO, $\pmb{1.06}\pm\pmb{0.08\%}. \pmb{1.13}\pm\pmb{0.08\%}$ and $\pmb{3.00}\pm\pmb{0.87\%}$ for SVR, respectively. These findings provide further evidence that footwear-based systems may represent valid alternatives to laboratory equipment for assessing a basic set of gait parameters in unconstrained environments.
Quantitative gait assessment typically involves optical motion capture systems and force plates, which result in high operating costs. Footwear-based motion tracking systems can provide a portable and affordable solution for real-time gait analysis in unconstrained environments. However, the relatively low accuracy of these systems still represents a barrier to their widespread use. In this paper, we show that linear and learning-based regression models can substantially improve the raw estimates of a set of kinematic gait parameters obtained with instrumented insoles (SportSole) from a group of $\mathbf{N}=\pmb{9}$ healthy subjects who walked at different speeds. Least Absolute Shrinkage and Selection Operator (LASSO) and Support Vector Regression (SVR) models are compared in terms of accuracy, precision, and robustness to change in gait speed, using gold-standard equipment to generate reference data. Results indicate that SVR is superior to LASSO. Indeed, the mean absolute errors (MAE) in stride length, velocity and foot-ground clearance were $\pmb{1.28}\pm\pmb{0.19\%}. \pmb{1.62}\pm\pmb{0.42\%}$ and $\pmb{3.72}\pm \pmb{0.87\%}$ for LASSO, $\pmb{1.06}\pm\pmb{0.08\%}. \pmb{1.13}\pm\pmb{0.08\%}$ and $\pmb{3.00}\pm\pmb{0.87\%}$ for SVR, respectively. These findings provide further evidence that footwear-based systems may represent valid alternatives to laboratory equipment for assessing a basic set of gait parameters in unconstrained environments.